AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Unifying Stable Optimization and Reference Regularization in RLHF

Li He, Qiang Qu, He Zhao, Stephen Wan, Dadong Wang, Lina Yao, Tongliang Liu

ICLR 2026 · Reinforcement learning › RL from feedback · approach: Theory of RLHF · 1 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add dual-KL penalties (pi_0 and pi_t) to advantage (ablation)PPO · Helpfulnessmean reward: 9.070 -> 9.522Improved
add dual-KL penalties with ratio clipping (ablation)PPO · Helpfulnessmean reward: 9.070 -> 9.581Improved
add ratio clipping to Dual-PPO (ablation)Dual-PPO · Helpfulnesswin rate over pi_0: 88.40% -> 90.37%Improved
replace PPO with DARPPO · TL;DR (Qwen2)reference win rate: 72.87% -> 98.27%Improved
replace GRPO with DARGRPO · Mean win rate (Qwen2)reference win rate: 85.15% -> 92.42%Improved
replace GRPO with DARGRPO · Harmlessness (Qwen2)reference win rate: 85.50% -> 85.84%No clear effect
replace RLOO with DARRLOO · Mean win rate (Qwen2.5)reference win rate: 76.65% -> 82.60%Improved
replace online DPO with DARonline DPO · Mean win rate (Qwen2)reference win rate: 83.63% -> 92.42%Improved
use online data instead of offline preference dataDPO · Mean win rate (Qwen2)reference win rate: 75.47% -> 83.63%Improved
replace GRPO with DARGRPO · MT-Bench GPT-4score: 8.425 -> 8.538Improved
replace RLOO with DARRLOO · AlpacaEval 2.0LC win rate: 52.25 -> 54.17Improved
use RL-based dual-KL (DAO) instead of regression (DAR) (ablation)DAR · Helpfulness and TL;DR reward curvestraining stabilityHurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy